基于神经网络的自适应预失真技术研究  

Research of Adaptive Digital Predistortion Based on Neural Network

作　　者：邱炜[1] 徐以涛[1] 任国春[1] 钟志明[1] 

机构地区：[1]解放军理工大学通信工程学院,江苏南京210007

出　　处：《现代电子技术》2009年第10期133-136,139,共5页Modern Electronics Technique

摘　　要：由于放大器固有的非线性特性,通信时会产生带内失真和临道干扰等问题,会对通信系统性能造成严重影响。为了克服放大器非线性特性,需要进行线性化处理。首先对功率放大器的非线性失真进行数学分析,简单介绍预失真技术的基本原理;其次在简单介绍神经网络的基础上提出一种简单的单入单出式神经网络自适应数字基带预失真技术。该技术能很好地改善三阶互调和五阶互调分量,且与一般多项式拟合预失真技术相比,在收敛速度和硬件实现难易程度上都有一定的优势。最后以双音信号为例进行了Matlab仿真。仿真结果印证了该技术的优势。Because of inherent nonlinearity of high power amplifier, f- band distortion and adjacent - channel interference, which may have bad influence on communication system. To overcome it, a linearization process is given. First, this paper makes a math analysis about nonlinearity distortion of HPA,deseribes the basic principle about digital predistortion. And then, after depicting simple neural network,the paper proposes an adaptive digital predistortion technology based on SISO- neural network, which can improve third -order and fifth -order intermodulation and comparing with polynomial - fitting technology, it has more advantage on convergence speed and difficulty of hardware realization. At last, the paper uses a double- sound signal to make a matlah simulation,the results also prove this technology＇s superiority.

关 键 词：功率放大器 预失真 神经网络 互调失真 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]